Jailbreak AI with Poetry | ASMR educational soft spoken
Publicado hace 2 semanas
Can a simple poem completely bypass advanced AI safety? In this soft-spoken educational ASMR video, we explore how researchers are using creative writing, metaphors, and poetic prompts to jailbreak large language models.
We'll look at a new study showing that adversarial poetry has a high ASR with jailbreaking AI in a variety of risk domains. In addition, we read a “Goblin Market” and “The Poison Tree,” both poems that utilize the same falsely euphonic tone to hide darker intentions and meanings.
I hope you enjoy ❤️
Love,
Jessa Lynn
#aisecurity #softspoken #unintentionalasmr
Creado:
Última actualización:
Última actualización:
